Optimizing M-estimation loss function

Background: The RoME framework utilizes M-estimation with a user-specified loss function (e.g., LS, LAD, Huber) to reconcile base forecasts, offering robustness over purely least-squares approaches like MinT. Selecting the appropriate loss function and tuning its parameters is crucial for optimal performance.

Question / Future Work: Selecting an optimal loss function and tuning its associated parameters within the M-estimation framework for RoME remains a challenging and application-dependent task. A promising avenue for further improvement involves systematically investigating advanced combination strategies for reconciled forecasts obtained from different loss functions, moving beyond simple averaging strategies.